-----GILBERT HARMAN------LANGUAGE, 1'HOUCH'I', AND COMMUNlCATION communication is the communication of thought. The parties involved typically communicate with the language they use in thinking. 111e words used to communicate a thought are the same as or similar to Lang;uage, Thought, and Communication those one "says to oneself" when one has that thought. Linguistic com­ munication does not typically require any complicated system of coding and decoding. Our usual translation scheme for understanding others is what Quine calls the homophonic one. Words are used to communicate thoughts that would ordinarily be thought in those or similar words. It I shall discuss two apparently conflicting views about our use of is true that allowance must sometimes be made for irony and other such natural language. The first view, that language is used primarily in devices; but in that case the thought communicated is some simple thought, has rarely been given explicit formulation but may be asso­ function of what would be normally communicated by a literal use of ciated with the theories of W . V. Quine and Wilfrid Sellars. The sec­ those words. ond view, that language is used primarily in communication, has been More precisely, linguistic communication typically involves commu­ explicitly put forward by Noam Chomsky, J. A. Fodor, and J. J. Katz, nication of what is sometimes called "propositional content." A speaker among others, and may also be associated (I think) with the theories says, "The door is shut," 'Shut the door," "Is the door shut?" or some of Paul Ziff and Donald Davidson. I shall describe each view and then such thing. He does so in part to get his hearer to think of the door's try to say where I think the truth lies. being shut. This first view holds that in such a situation, if communica­ 1. The view that language is used primarily in thought. This view is tion is successful, the hearer will think of the door's being shut by not that all or even most thinking or theorizing is in some natural lan­ adopting the appropriate sentential attitude. We might say that the guage. We may reasonably suppose that animals think, that children hearer attends to the sentence, "The door is shut," where this is a tech­ can think before they learn a natural language, and that speakers of a nical sense of "attends to." natural language can have thoughts they cannot express in language. Notice that the claim is not that a person can think of the door's The view is rather that anyone who fully learns a natural language can being shut only by attending to the sentence, "The door is shut." He and does sometimes think in that language. More precisely, it is that might instead attend to an "equivalent" sentence, where the relevant some of a speaker's so-called propositional attitudes are to be construed sort of equivalence is that discussed near the end of this paper; he as, at bottom, attitudes toward sentences of his language. A speaker of might even attend to a nonlinguistic representation that is part of a English may believe that the door is open by believing-true the sen­ system of representation he uses in thinking, as long as the representa­ tence, "The door is open." Another may fear that the door is open by tion is relevantly equivalent to "The door is shut" in English. The claim fearing-true "The door is open." A third may think of the door's being is rather that normally, when a speaker successfully communicates in open by adopting the appropriate attitude toward "The door is open." English by saying, "The door is shut," etc., the hearer thinks of the Strictly speaking, sentential attitudes involve sentences conceived un­ door's being shut by attending to the English sentence, "The door is der one or another, more or less detailed grammatical analysis. I shall shut." There will normally be a relatively simple grammatical relation­ return to this point near the end of this essay. For now, I shall speak ship between the sentence the speaker uses to communicate certain loosely of sentential attitudes as attitudes toward sentences. "propositional content" and the sentence to which the hearer must at­ The view that the primary use of language is in thought has roughly tend if communication is to be successful. The "deep structure" of the the following implications for the theory of communication. Linguistic latter sentence is the same as or a part of the deep structure of the former. Since sentential attitudes involve sentences conceived under AUTH®R's NOTE: The preparation of this paper was supported in part by grants from NEH and NSF. grammatical analyses, it is sufficient that the hearer should attend to 270 271 Gilbert Harman l.ANCUAGE, THOUCllT, AND COMMUNICATION the sentence uttered conceived under the appropriate analysis. In this or ideas and the decoding of the phonetic and syntactic structure ex­ sense, linguistic communication does not typically make use of compli­ hibited in such a physical phenomenon by other speakers in the form 1 cated principles of coding and decoding and our usual translation of an inner, private experience of the same thoughts or ideas. scheme is the homophonic one. The hearer need only hear the sentence Katz takes seriously the notion that linguistic communication ordinarily uttered as having the appropriate structural description. He does not and typically involves such coding and decoding. He does not agree that need to go on to translate the sentence, under that description, into typically one thinks in language and that the most usual "code" used anything else in order to understand it. in communication is the homophonic one. He does not believe that the Proponents of the view that language is primarily used in thought can words used in communication are usually the same as or similar to those point out that, although one might use a natural language as a code, that make up the thought communicated. As a result he sees linguistic so that one's listeners would have to use complicated principles of de­ communication as a relatively complicated business and he takes the coding in order to understand what has been said, this would not be main use of language to be its use in communication: an ordinary case of linguistic communication. They would also point Natural languages are vehicles for communication in which syntac­ out that, when a person learns a second language, he may at first have tically structured and acoustically realized objects transmit meaningful to treat the new language as a code; but hopefully he soon learns to messages from one speaker to the other. . . . The basic question that think directly in the second language and to communicate with other can be asked about natural languages is: what are the principles for re­ lating acoustic objects to meaningful messages that make a natural lan­ speakers of· that language in the ordinary way, which does not involve guage so important and flexible a form of communication? 2 complex coding and decoding or any sort of translation. Furthermore, proponents can say, when a person thinks out loud, it I think that in theory Noam Chomsky rejects Katz's view of commu­ is not always true that he has to find a linguistic way to express some­ nication and natural languages. He takes the primary function of lan­ thing that exists apart from language. Without language many thoughts guage to be its use in the free expression of thought. He speaks approv­ and other propositional attitudes would not even be possible. In learn­ ingly of Humboldt's emphasis on the connection between language and ing his first natural language, a child does not simply learn a code thought, especially the way in which a particular language brings with which he can use in communicating his thoughts to others and in de­ it a world view that colors perception, thought, and feeling. According coding what they say. He acquires a system of representation in which to Chomsky's description of Humboldt's view, to have a language is to he may express thoughts made possible by that very system. This is ob­ have a system of concepts vious when one acquires for the first time the language of a science or and it is the place of a concept within this system (which may differ of mathematics. The claim is that it is no less true when one learns his somewhat from speaker to speaker) that, in part, determines the way in which the hearer understands a linguistic expression . . . [T]he con­ first natural language. cepts so formed are systematically interrelated in an "inner totality," That provides a rough sketch of the view that language is used pri­ with varying interconnections and structural relations ... This inner marily in thought. I shall say more about that view below. Now I want totality, formed by the use of language in thought, conception, and ex­ to describe the apparently conflicting theory that communication pro­ pression of feeling, functions as a conceptual world interposed through vides the primary use of language. the constant activity of the mind between itself and the actual objects, and it is within this system that a word obtains its value . . . Conse­ 2. The view that language is used primarily in communication. J. J. quently, a language should not be regarded merely, or primarily, as a Katz states explicitly a view that is implicit in many things said by other means of communication . . . and the instrumental use of language linguists: 1 J. J. Katz, The Philosophy of Language (New York: Harper and Row, 1966), Roughly, linguistic communication consists in the production of p. 98. some external, publicly observable, acoustic phenomenon whose pho­ 2 Ibid. Cf. J. J. Katz, "Recent Issues in Semantic Theory," Foundations of Lan­ netic and syntactic structure encodes a speaker's inner, private thoughts guage, 3(1967) :125. 272 273 Gilbert Harman J, ANCUACI~ , TllOUCHT, AND COMMUNICATION (its use for achieving concrete aims) is derivative and subsidiary. It is, the view that speakers have (unconscious) knowledge of the rules of gram­ for Humboldt, typical only of parasitic systems (e.g .... the lingua mar. In a paper presented to the 1968 New York University Institute of 3 franca along the Mediterranean coast) . Philosophy7 Chomsky surveys criticism of his views and denies they are Chomsky's approval of Humboldt suggests that he would accept the based on the above confusions. He acknowledges that it would be "ab­ view that language is used primarily in thought and would reject Katz's surd" to suppose a typical speaker to have knowledge that the language remarks about language and communication. But other speculations of is described by certain rules of grammar and claims he wants to say only Chomsky's seem to make sense only if he does at least unconsciously that a speaker has knowledge of the rules of grammar. On the other hand, accept a view like Katz's, only if he does see communication as the there and elsewhere Chomsky describes the speaker's knowledge of the central function of language, only if he does think linguistics is prima­ rules as "a system of beliefs" and as a "theory"; and he says that a child rily concerned with the speaker-hearer rather than the thinker, and only who acquires knowledge of the rules "determines that the structure of if he does take communication to involve a complicated process of de­ his language has the specific characteristics that empirical investigation of coding. language leads us to postulate . . ." 8 Surely it will be interesting to see Consider the view, held by both Katz and Chomsky, that speakers of what has led Chomsky to such a position. a language have (unconscious) knowledge of the grammatical rules of Simple common sense can account for his denial that a typical speak­ the language. We would not ordinarily say that a typical speaker of er of a language has unconscious knowledge that the language is de­ English has knowledge of the rules of a transformational grammar of scribed by certain rules of grammar. But it is difficult to say what leads English. We would ordinarily attribute such knowledge to a gram­ him simultaneously to assert that the speaker has knowledge of the marian if to anyone. It has not been easy to discover why Chomsky and rules, knowledge which consists of a system of beliefs or of a theory Katz attribute such knowledge to speakers generally. I and others have that the speaker developed in the course of determining that those rules hypothesized that Chomsky and Katz confuse knowing how to use a describe the language. Chomsky denies that he makes the mistakes sug­ language that is described by the grammatical rules with knowing that gested above. We can only take him at his word. Therefore we must the grammar is described by those rules.4 I have also suggested that look elsewhere for an explanation of his view. Chomsky may be confusing "knowing that certain sentences are gram­ I suggest we need look no farther than Katz's book on the philosophy matically unacceptable, ambiguous, etc., with knowing the rules of of language. It is easy to see why he attributes to speakers unconscious grammar by virtue of which sentences are unacceptable, ambiguous, knowledge of the rules of grammar. He takes linguistic communication etc." 5 I have also suggested that Chomsky fails "to distinguish princi­ to involve complex coding and decoding. Since such communication ples that account for how a person's mind works from principles a per­ typically involves novel messages not previously encountered, he thinks son accepts as true." 6 speakers and hearers will need to know rules that relate sequences of Although these suggestions may contain part of the truth, they do code to messages. Such rules are provided by transformational genera­ not account for the tenacity with which Chomsky (for one) holds on to tive grammars that define the sound-meaning relationship for the lan­ guage in question. Therefore he thinks that speakers and hearers must •Noam Chomsky, "Current Issues in Linguistic Theory," in J. A. Fodor and J. J. Katz, eds., The Structure of Language (Englewood Cliffs, N.J.: Prentice-Hall, have knowledge of rules that are equivalent to the rules of an adequate 1964), pp. 58-59. See also Chomsky's remarks about Humboldt in his book Car­ transformational generative grammar of the language. tesian Linguistics (New York: Harper and Row, 1966). •Three authors endorse this suggestion in a single issue of Journal of Philosophy, . • "Linguistics and Philosophy," in Language and Philosophy, ed. Sidney Hook 64 ( 1967) : N. L. Wilson, "Linguistic Butter and Philosophical Parsnips," pp. (New York : New York University Press, 1969), pp. 51-94. 55-67; Henry Hiz, "Methodological Aspects of the Theory of Syntax," pp. 67-84; •This last remark is from Chomsky's part in a symposium on "Recent Contribu· and Gilbert Harman, "Psychological Aspects of the Theory of Syntax," pp. 75-87. tions to the Theory of Innate Ideas" (Boston Studies in the Philosophy of Science, •Harman, "Psychological Aspects of the Theory of Syntax," p. 82. vol. 3, p. 86). Most of Chomsky's N.Y.U. paper is devoted to clarifying that earlier •Harman, review of Chomsky's Cartesian Linguistics in Philosophical Review, paper and responding to remarks in that symposium by Putnam and by Goodman. 72 (1968) :234. (The emphasis on that is mine.) 274 275 Gilbert Harman I..ANCUAGI;: , TIJOUCIJT, AND OMMUNJCATION

To understand the ability of natural languages to serve as instruments whi ch :1 pl1011ctic representation equivalent to the one into which the for the communication of thoughts and ideas we must understand what speaker encoded his message is obtained. This represen tation is decided .it is that permits those who speak them consistently to connect the into a representation of the same message that the speaker originally right sounds with the right meanings. chose to convey by the hearer's equivalent system of linguistic rules. It is quite clear that, in some sense, one who knows a natural lan­ Hence, because the hearer employs the same system of rules to decode guage tacitly knows a system of rules. This is the only assumption by that the speaker employs to encode, an instance of successful linguistic which we can account for a speaker's impressive ability to use language communication occurs .... creatively. Fluent speakers both produce and understand sentences that . . . A linguistic description . . . describes the knowledge whose pos­ they have never previously encountered, and they can do this for in­ session permits a fluent speaker to communicate with other speakers of definitely many such novel sentences.9 his language L and whose absence prevents those who only speak an­ other language from communicating with normal monolingual speakers Given that fluent speakers are fluent because of their knowledge of of Lin their language.10 the rnles of the language and that linguistic communication is a process in which the meaning that a speaker connects with the sounds he utters Chomsky says nothing quite as explicit; but a similar view of linguis­ is the same meaning that the hearer connects with these same sounds, tic communication seems to lie behind his version of the thesis that it seems necessary to conclude that speakers of a natural language com­ speakers of a language have unconscious knowledge of the grammatical municate with each other in their language because each possesses es­ rules of that language. If Chomsky sees communication as requiring sentially the same system of rules. Communication can take place be­ cause a speaker encodes a message using the same linguistic rules that the hearer to decode a linguistic message into the appropriate nonverbal his hearer uses to decode it. This becomes clearer when we think of thought, then it would be natural for him to believe that a "speaker­ how we learn a foreign language in the classroom. Our teacher and text hearer" must know the appropriate code, must know the rules for cod­ present us with a more or less accurate approximation of the rules that ing and decoding. The speaker has never been explicitly taught the any ~peaker of the foreign language tacitly knows. Our task is to learn code, so he must have developed a theory about it, a system of beliefs them well enough for us to produce utterances that can be decoded by speakers of that foreign language and to understand utterances of those about it, on the basis of his observation of other speaker-hearers. He speakers themselves. This sort of example brings out the fact that our must have determined that the rules are whatever they are. Since rules competence in a foreign language depends on whether, and to what linking sound and meaning are what the grammar provides, the speaker­ extent, the rules we have been taught are equivalent to those that speak­ hearer has unconscious knowledge of the rules of grammar. ers ·Of the foreign language acquired naturally. But it also shows that Some such line of thought is responsible for Chomsky's view of lin­ each speaker of the foreign language must use essentially the same sys­ tem of rules . . . guistic competence. That he does not go on to say (in fact he explicitly . . . RougbJy, and somewhat metaphorically, we can say that some­ denies) that the speaker has unconscious knowledge that the rules are thing of the following sort goes on when successful linguistic commu­ whatever they are must be chalked up to the sudden intrusion of com­ nication takes place. The speaker, for reasons that are linguistically ir­ mon sense plus the fact that the above picture of communication is relevant, chooses some message he wants to convey to his listeners: not his explicit view but only his tacit background assumption. some thought he wants them to receive or some command he wants to .give them or some question he wants to ask. This message is encoded Let me try to document this. In Chomsky's explicit remarks about in the form of a phonetic representation of an utterance by means of the relations between thought, language, and communication, he en­ the system of linguistic rules with which the speaker is equipped. This dorses the Humboldtian view which is on the whole an elaboration of encoding them becomes a signal to the speaker's articulatory organs, and part of the first sort of view, which I described at the beginning of this he vocalizes an utterance of the proper phonetic shape. This is, in turn, essay. Language is primarily an instrument for the free expression of picked up by the hearer's auditory organs. The speech sounds that stimulate these organs are then converted into a neural signal from 10 Ibid., pp. 102-105. Cf. J. J. Katz, "Mentalism in Linguistics," Language, 40 ( 1964) : 124-137; reprinted in Leon A. Jakobovitz and Murray S. Miron, eds., •Katz, The Philosophy of Language, p. 100. Readings in the Psychology of Language (Englewood Cliffs, N.J.: Prentice-Hall, 1967). 276 277 Gilbert Harman LANCUACJ!:, THOUCJIT, AND COMMUNICATION thought; the instrumental use of language, e.g., in communication, is The grammar as a whole can thus be regarded, ultimately, as a device secondary and based on its primary use; etc.11 In practice, however, for pairing phonetically represented signals with semantic interpreta­ Chomsky appears to take a different line: "The central fact to which tions . . . . In performing as a speaker or hearer, [one] puts this [gram­ mar] to use. Thus as a hearer, his problem is to determine the struc­ any significant linguistic theory must address itself is this: a mature tural description assigned by his grammar to a presented utterance (or, speaker can produce a new sentence of his language on the appropriate where the sentence is syntactically ambiguous, to determine the correct occasion, and other speakers can understand it immediately, though it structural description for this particular token), and using the informa­ is equally new to them." 12 He begins his book, Aspects of the Theory tion in the structural description, to understand the utterance.16 of Syntax, with the remark, "Linguistic theory is concerned primarily It is quite obvious that sentences have an intrinsic meaning determined with an ideal speaker-listener . . ." 13 Similarly, he begins a recent sur­ by linguistic rule and that a person with command of a language has vey article by saying, "At the crudest level of description, we may say in some way internalized the system of rules that determine both the that a language associates sound and meaning in a particular way; to phonetic shape of the sentence and its intrinsic semantic content - that he has developed what we will refer to as a specific linguistic com­ have command of a language is to be able, in principle, to understand petence.... [T)he technical term 'competence' refers to the ability what is said and to produce a signal with an intended semantic inter­ of the idealized speaker-hearer to associate sounds and meanings strictly pretation." 14 In these and many similar passages Humboldt seems to in accordance with the rules of his language. The grammar of a lan­ have been forgotten. Language is no longer seen primarily as an instru­ guage, as a model for idealized competence, establishes a certain rela­ ment for the free expression of thought but as an instrument for the tion between sound and meaning - between phonetic and semantic representations ... Clearly, in understanding a signal, a hearer brings communication of thought. to bear information about the structure of his language.17 I think that Chomsky must be taking the thought communicated to be nonverbal. Otherwise he would not be so impressed by what he Chomsky and Halle put the point this way: calls the "creativity of language": "The most striking aspect of linguis­ The person who has acquired knowledge of a language has internalized tic competence is what we may call the 'creativity of language', that a system of rules that determines sound-meaning connections for in­ definitely many sentences. Of course, the person who knows a language is, the speaker's ability to produce new sentences, sentences that are perfectly has little or no conscious knowledge of the rules that he uses immediately understood by other speakers although they bear no physi­ constantly in speaking or hearing, writing or reading, or internal mono­ cal resemblance to sentences which are 'familiar'.15 This aspect of lin­ logue. It is this system of rules that enables him to produce and inter­ guistic competence will only seem "striking" to someone who does not pret sentences that he has never before encountered.18 think that the relevant thoughts are in words and that the relevant prin­ What I have been trying to show is that Chomsky's and Katz's talk ciple for interpreting what others say is to use the homophonic system about unconscious knowledge of the rules of grammar represents more of translation from words said to words thought. It seems to me that than a careless use of the words "know" and "knowledge." More is in­ Chomsky's explanation of this "striking aspect of linguistic compe­ volved than a confusion between knowing-that and knowing-how, be­ tence" is like Katz's: The speaker knows what the sound-meaning cor­ tween knowing that a sentence is ambiguous and knowing the gram­ respondence for the language is, he knows the rules of grammar that matical rules by virtue of which it is ambiguous, or between principles specify this correspondence. Chomsky has put his claim this way: that account for knowledge of a language and knowledge of those prin­ 11 See above, pp. 273-74 and footnote 3. ciples. Their talk about unconscious knowledge of the rules of grammar 12 Chomsky, "Current Issues in Linguistic Theory," p. 50; my emphasis. 1 11 P. 3. • Chomsky, "Current Issues in Linguistic Theory," p. 52 . 1 • Chomsky, "The Formal Nature of Language," pp. 397-399. "Chomsky, "The Formal Nature of Language," Appendix A to Eric H. Lenne­ 1 berg, Biological Foundations of Language (New York: Wiley, 1967), p. 397. • Noam Chomsky and Morris Halle, The Sound Pattern of English (New York: 10 Chomsky, Topics in the Theory of Generative Grammar (The Hague: Mouton, Harper and Row, 1968) , p. 3. Cf. Chomsky, "Recent Contributions to the Theory 1966) , p. 11. Cf. Chomsky, "Recent Contributions to the Theory of Innate Ideas," of Innate Ideas," pp. 83-84. pp. 82-83. 279 278 Gilbert Harman LANCUACF., TllOUCHT, AND COMMUNICATION ultimately reflects a basic conception of our use of language and a defi­ component provides a phonetic shape for a sentence, the semantic com­ nite picture of linguistic communication. ponent provides a representation of that message which actual utter­ Let me note in passing that we can now understand Chomsky's and ances haying this phonetic shape convey to speakers of the language in Katz's otherwise puzzling acceptance of what they call a theory of "in­ normal speech situations." 2° Katz adds that nate ideas." They think of the child about to learn a language as faced The semantic component . . . must contain rules that provide a mean­ with a gigantic cryptogram, a code-breaking problem of the toughest ing for every sentence generated by the syntactic component .... sort. They are impressed that children are able to break the code so These rules, therefore, explicate an ability to interpret infinitely many fast, without any training, all by themselves. Perhaps they think about sentences. . . . The hypothesis on which we will base our model of the semantic component is that the process by which a speaker inter­ how long it took a whole group of adults to break the Japanese code. prets each of the infinitely many sentences is a compositional process in In order to account for a child's performance they infer that he must which the meaning of any syntactically compound constituent of a have had information about the nature of the code ahead of time; and sentence is obtained as a function of the meanings of the parts of the they equate such innate information with what used to be called innate constituent. Hence, for the semantic component to reconstruct the ideas.19 In this case too, Chomsky and Katz have not simply misused principles underlying the speaker's semantic competence, the rules of the semantic component must simulate the operation of these princi­ terms like "empiricism," "rationalism," "innate ideas," etc. That's why ples by projecting representations of the meaning of higher level con­ complaints about their usage fail to move them. Their talk about innate stituents from representations of the meaning of the lower level constitu­ ideas is ultimately based on their conception of the use of language in ents that comprise them ... communication. This means that the semantic component will have two subcompo­ 3. Two views of semantics. What I will call a compositional theory of nents: a dictionary that provides a representation of the meaning of each of the words in the language, and a system of projection rules that meaning holds that a hearer determines what the meaning of an utter­ provide the combinatorial machinery for projecting the semantic repre­ ance is on the basis of his knowledge of the meaning of its parts and sentation for all supraword constituents in a sentence from the rep­ his knowledge of its syntactic structure. Such a view follows naturally resentations that are given in the dictionary for the meanings of the from the picture of communication that takes it to involve complex words in the sentence.21 coding and decoding. On that picture, to understand (the meaning of) The projection rules of the semantic component for a language char­ a sentence is to know what (nonverbal) thought or thoughts the sen­ acterize the meaning of all syntactically well-formed constituents of tence encodes. Meanings are identified with the relevant thoughts. two or more words on the basis of what the dictionary specifies about Hence the view is that to know the meaning of an expression is to these words. Thus, these rules provide a reconstruction of the process by which a speaker utilizes his knowledge of the dictionary to obtain know what meaning, i.e., thought, the code associates with that sen­ the meanings of any syntactically compound constituent, including sen- tence. On this view, a general theory of meaning of a language is given tences.22 · by the principles of the code that defines the sound-meaning corre­ Here meanings are to be identified with "readings." spondence for the language. Because of the unbounded nature of lan­ guage, these principles would have to be compositional or recursive. Projection rules operate on underlying phrase markers that are par­ tially interpreted in the sense of having sets of readings assigned only Recall that for transformational linguistics, the rules of coding and to the lower level elements in them. They combine readings already decoding are given by the grammar. This has three components, a syn­ assigned to constituents to form derived readings for constituents which, tactic component that connects a phonological component with a se­ as yet, have had no readings assigned to them . . . Each constituent of mantic component. And as Katz describes it, "whereas the phonological 20 Katz, The Philosophy of Language, p. 151. 21 19 Chomsky, Aspects of the Theory of Syntax, pp. 25-59; The Philosophy of Ibid., pp. 152-153. Cf. J. A. Fodor, "Some Remarks on the Philosophy of Lan­ Language, pp. 240-282. See also the papers by Chomsky and responses to them in guage," in F. H. Donell, ed., Aspects of Contemporary American Philosophy (Wurz­ the works cited in footnotes 7 and 8 above. burg-Wien : Physica-Verlag, 1965), pp. 82-83. .. Ibid., pp. 161-162. 280 281 Gilbert Harman LANGUAGE, THOUCIIT, AND COMMUNICATION

a~ underlying. phrase marker is thus assigned a set of readings, until the linguistic rule." 24 ) '111at is exactly the sort of view Bloomficld 2 ~ at­ highest constituent, the whole sentence, is reached and assigned a set tacked as "mentalistic." In an influential article,26 Katz attempts to an­ of readings, too.2a swer Bloomfield but only blurs the issue, since he never considers It is easy to begin to see what is going on here. This theory would whether Bloomfield's criticisms of mentalism apply against a theory certainly be appropriate as an account of the meaning of expressions which assumes that semantics can be unconcerned with the meaning of in one language, e.g., Russian, given in another language taken to be thoughts because of their supposed intrinsic intelligibility. antecedently understood, e.g., English. In that case one would want Thus, distinguish a theory of the meaning of thoughts from a theory some general principles for translating Russian into English. Such prin­ of the meaning of messages. I would argue that a theory of the mean­ ciples would enable one to know the meaning of Russian expressions ing of messages presupposes a theory of the meaning of thoughts.27 The because one already knows (in the ordinary sense of this phrase) the former theory might resemble that proposed by Paul Grice.28 According meaning of the corresponding English expressions. Katz tries to make to Grice, the thought meant is the one the speaker intends the hearer the same trick work in order to give an account of the meaning of sen­ to think the speaker has, by virtue of his recognition of the speaker's tences in English. In effect he envisions a theory that gives principles intention. However that may be, in normal linguistic communication, for translating from English into Mentalese. He thinks these principles speakers and hearers rely on a regular association between messages and are sufficient because one already knows Mentalese. thoughts. Those philosophers and linguists who think that language is Proponents of the view that language is primarily used for thought used primarily in communication suppose that this association is be­ will raise two objections to Katz's theory of communication. They will tween sentences in, e.g., English (conceived under particular structural argue first that the theory is circular, since (as they maintain) Mental­ descriptions) and, as it were, sentences in Mentalese. Those who be­ ese is simply English used to think in. Second, they will claim that, lieve that language is used primarily in thought suppose that the asso­ even apart from that, Katz's maneuver simply shifts the problem back ciation is a trivial one, since the language used to communicate with one step. For what would it be to give an account of meaning for Men­ is normally the same as that used to think with. talese? One cannot continue forever to give as one's theory of meaning A theory of the meaning of thought is either a theory of meaning for a way to translate one system of representation into another. At some natural languages (as is claimed by those who take language to be used point a different account is needed. Katz's theory only delays the mo­ primarily in thought) or a theory of Mentalese (if those philosophers ment of confrontation. and linguists are right who take natural language to be used primarily in There are two distinct issues here. The first is the difficult quasi­ communication). In either case, it must exploit the Humboldtian in­ empirical question whether thought is in the relevant sense verbal. I sight that the meaning of a linguistic expression is derived from its func­ shall postpone discussion of that issue. The second is the methodologi­ tion in thought as determined by its place in one's total conceptual cal question whether a semantic theory may presuppose a theory of the scheme.29 One must consider the influence of perception on thought, nature of thought. Proponents of the view that language is primarily the role of inference in allowing one to pass from some thoughts to used for thought take semantics to be part of such a theory of the na­ others, and the way thought leads to action. The theory we want will ture of thought. They argue, as it were, that semantics must be con­ "'Chomsky, "The Formal Nature of Language," p. 397. cerned in the first instance with the meaning of thoughts. 25 Leonard Bloomfield, Language (New York: Holt, 1933), and "Linguistic As­ Katz seems to assume that no account need be given of the meaning pects of Science," International Encyclopedia of Unified Science, vol. 1 (Chicago: University of Chicago Press, 19 55 ) . of thoughts, as if Mentalese were intrinsically intelligible. (Cf. Chom­ 111 "Mentalism in Linguistics." sky's remark that "sentences have an intrinsic meaning determined by "'See my "Three Levels of Meaning/' Journal of Philosophy, 65 ( 1968): 590-602. '"Paul Grice, "Meaning," Philosophical Review, 66 ( 1957): 377'-388; "Utterer's "'Ibid., pp. 164-165. Meaning and Intentions," Philosophical Review, 78 ( 1969) :147-177. '"'See above, pp. 273-74 and footnote 3. 282 283 GillJcrt Harman LAN CUACE, 'l'llOUGllT, AND COMMUNICATION be like that proposed by Wilfrid Sellars in his paper on "language described above, which ass umes that to give an account of mea ning for 30 games." Sellars identifies the meaning of an expression with its (po­ some language is to say how a "speaker-hearer" is able to correlate mean­ tential) role in the evidence-inference-action language game of thought. ings, qµa thoughts, with sentences. Our suspicions are confirmed by Similar theories have been proposed by various philosophers, e.g., Car­ Davidson's explicit argument for the compositional theory: 31 2 nap, Ayer, and Hampshire. And Quine3 has argued that meaning at These matters appear to be connected in the following informal way this level admits of a special sort of indeterminacy. with the possibility of learning a language. When we can regard the According to proponents of the view that language is used primarily meaning of each sentence as a function of a finite number of features in thought, transformational semantics, with its compositional theory of the sentence, we have an insight not only into what there is to be of meaning, provides neither an account of the meaning of language learned; we also understand how an infinite aptitude can be encom­ passed by finite accomplishments. Suppose on the other hand the lan­ as used to · think with nor an account of the meaning of language as guage lacks this feature; then no matter how many sentences a would-be used to communicate with. They claim that it cannot provide a theory speaker learns to produce and understand, there will remain others of the meaning of thought, since a speaker does not understand the whose meanings are not given by the rules already mastered. It is natu­ words he uses in thinking by assigning readings to them, and that it ral to say such a language is unlearnable. This argument depends, of cannot provide a theory of the meaning of a message, since it treats a course, on a number of empirical assumptions: for example, that we do not at some point suddenly acquire an ability to intuit the meanings relatively simple problem of interpretation as if it were quite compli­ of sentences on no rule at all; that each new item of vocabulary, or new cated. grammatical rule takes some finite time to be learned; that man is mor­ 4. Composition and communication. Compositional theories of mean­ taJ.35 ing depend on the view that language is used primarily in communica­ This argument makes sense only in the presence of an assumption, tion. Thus Davidson argues as follows. He says, first, "we are entitled which Davidson explicitly acknowledges elsewhere, that "speakers of a to consider in advance of empirical study what we shall count as know­ language can effectively determine the meaning or meanings of an arbi­ ing a language, how we shall describe the skill or ability of a person trary expression (if it has a meaning)," 36 where that is understood to who has learned to speak a language." 33 He wants to argue for the con­ mean that a speaker (hearer) understands a sentence by translating it dition "that we must be able to specify, in a way that depends effec­ into its Mentalese counterpart (and where Mentalese is not the lan­ tively and solely on formal considerations, what every sentence means. guage used in communication). If speakers of a language can effectively With the right psychological trappings, our theory should equip us to determine the meaning of messages in ordinary linguistic communica­ say, for an arbitrary sentence, what a speaker of the language means by tion by using the homophonic mapping of verbal message onto verbal that sentence (or takes it to mean) ." 34 That last reference, to what a thought, the assumption does not support Davidson's argument for a speaker takes the sentence to mean, sounds suspiciously like Katz's view compositional theory of meaning. In the end Davidson argues for a version of the theory that meaning 30 Wilfrid Sellars, "Some Reflections on Language Games " in Science Perception and Reality (London : Routledge and Kegan Paul, 1963) . ' ' ' is given by truth conditions.37 That is no real improvement over Katz's 81 E.g., Rudolf Carnap;, "Tes.tability and Meaning," Philosophy of Science, 3 theory, from the point of view of those who take language to be used (193?) and ~ (1937 ); Meanmg and Synonymy in Natural Languages," Philo­ sop~1cal Studies, 7 (1955) :33-47. A. J. Ayer, Language, Truth, and Logic (1936) , primarily in thought. According to Katz, a speaker knows the meaning repnnt.ed as a Dover paperback; The Foundations of Empirical Knowledge (London: of sentences of his language because he has mastered the complicated Macmillan, 1940 ). Stuart Hampshire, Thought and Action (London: Chatto and Windus, 19 59). rules (of a transformational generative grammar of his language) that cor­ :w.V. Quin~, Wo~?and <;'>bject (Can;ibridge, Mass .: M.I.T. Press, 1960) . relate sentences with thoughts. According to Davidson, a speaker knows . Donald Day1dson, Theories of Meanmg and Leamable Languages," in Y. Bar­ H11!el, ed., Logic, Methodology, and Philosophy of Science: Proceedings of the 1964 ""Ibid., pp. 387-388. International Congress (Amsterdam: North-Holland 1965) p. 387. .. Donald Davidson, "Truth and Meaning," Synthese, 17 (1967) :320. 34 Ibid. ' ' "'Ibid., pp. 304-323. 284 285 Gilbert Harman LANGUAGE, THOUGHT, AND COMMUNICATION the meaning of sentences of his language because he has mastered the vant nonsyntactic semantic regularities. complicated rules (of a truth definition for his language) that correlate Ziff says, "Jn formulating the theory presented here [in Semantic sentences with truth conditions. Katz's theory is in trouble if the rele­ Analysis] J have had but one objective in mind, viz. that of determining vant thoughts are verbal. The same difficulty faces Davidson in a slight­ a method and a means of evaluating and choosing between competing ly different form, if the relevant knowledge of truth conditions will analyses of words and utterances." 41 If "analysis" here means "philo­ be verbal. Katz would say that the speaker understands the sentence sophical analysis," Ziff's enterprise must be counted a success, especially "Snow is white" by virtue of the fact that he has correlated it with the in the light of his careful discussion of the analysis of the word "good" thought that snow is white. Davidson would (presumably) say that in the final chapter. And since analysis is perhaps a kind of translation the speaker understands that sentence by virtue of the fact that he or decoding, it may be possible to defend a compositional theory of knows it is true if and only if snow is white. The difficulty in either meaning as a compositional theory of analysis. (One must see that a case is that the speaker needs some way to represent to himself snow's proposed analysis of a word is adequate for various contexts and is con­ being white. If the relevant speaker uses the words "snow is white" to sistent with analyses suggested for other words.) represent in the relevant way that snow is white, both Katz's and Da­ But Ziff does not give quite that argument for a compositional theory vidson's theories would be circular. And, if speakers have available a qf meaning; and the argument he does give indicates that he wants form of Mentalese in which they can represent that snow is white, so more from his theory than a way of evaluating philosophical analyses. that the two theories avoid circularity, there is still the problem of His own argument seems to assume that a speaker understands sen­ meaning for Mentalese. tences by virtue of being able to give analyses or explications of them. The point is that no reason has been given for a compositionaltheory Jn Semantic Analysis the argument goes like this: of meaning for whatever system of representation we think in, be it Jn a general form, the principle of composition is absolutely essential Mentalese or English. This point has obvious implications for linguis­ to anything that we are prepared to call a natural language, a language tics, and for philosophy too if only of a negative sort. For e:Xample, that can be spoken and understood in the way any natural language can Davidson uses his theory in order to support objections to certain in fact be spoken or understood...... theories about the logical form of belief sentences.38 Since his argument How is it that one can understand what .is said if what is said has not for a compositional theory of meaning fails for the language one thinks been said before? Any language whatever allows for the utterance. of new utterances both by the reiteration of old ones and by the formation in, those objections have no force against theories about the logical form of new ones out of combinations of old elements. Hence any natural of belief sentences used in thinking or theorizing. language whatever allows for the utterance of both novel utterance tok­ Similar remarks apply to the compositional theory of meaning in Paul ens and novel utterance types. If a new utterance is uttered and if the Ziffs Semantic Analysis.39 He argues that "the semantic analysis of an utterance is not then and there to be given an arbitrary explication, that utterance consists in associating with it some set of conditions [and] one is able to understand what is said in or by uttering the utterance must in some way at least be partially owing to one's familiarity with that the semantic analysis of a morphological element having meaning the syntactic structure of the utterance.42 in the language consists in associating with it some set of conditions 40 Jn "On Understanding Understanding Utterances" Ziff is more explicit: •••" Very roughly speaking, the relevant conditions are those that must obtain if something is to be uttered without deviance from rele- part of what is involved in understanding an utterance is understanding what conditions are relevantly associated with the utterance. . . . . 88 See Davidson, "Theories of Meaning and Learnable Languages," and reference Someone says 'Hippopotami are grace~ul' ~nd .we understa~d what is therein. said. In some cases we understand what is said without attendmg to the .. Ithaca, N.Y.: Cornell University Press, 1960. '° Paul Ziff, "On Understanding Understanding Utterances," in Fodor and Katz, discourse the utterance has occurred in or without attending to the con- The Structure of Language, and in Ziff, Philosophic Turnings (Ithaca, N.Y.: Cornell " Semantic Analysis, sec. 201. University Press, 1966), sec. 3. .. Semantic Analysis, sec. 64. 286 287 Gilbert Harman LANGUAGE, 'J'llOUGIIT, AND COMMUNICATION text of utterance. How do we do it? posed to account for "the way that speakers understand sentences." 4r. It seems reasonable to suppose that part of what is involved is this: That suggests it is the sort of theory Grice has tried to develop. But the Such an utterance is understood on the basis of its syntactic structure authors, go on to describe it as an account of the meaning or meanings and morphemic constitution. Assuming that part of what is involved in understanding an utter­ a sentence has when taken in isplation from possible settings in actual ance is understanding what conditions are relevantly associated with the discourse. In other words their theory is restricted to giving an ac­ utterance, this means that we take a certain set of conditions to be count of meaning for those cases in which the message communicates a associated with such an utterance on the basis of its syntactic structure thought that (on the other view) can be expressed in (roughly) the 4 and morphemic constitution. 3 same words as those in which the message is expressed. They argue that To this the same remarks apply as to the theories of Katz and of Da­ another theory would have to account for the interpretation or inter­ vidson. A speaker can understand that certain conditions are associated pretations assigned when a sentence occurs in a particular setting. Fur­ with an utterance and can take certain conditions to be associated with thermore they argue that the latter theory must presuppose the one an utterance only if he has some way to represent to himself that the they present. Thus they come close to the other view's distinction be­ conditions are associated with the utterance. And even if the speaker tween levels one and two, and in a way they attempt to provide a uses Mentalese to represent utterance-conditions correlations, the prob­ theory of level one. Or rather, Katz and Fodor see the need for three lem of meaning is merely pushed back one step to Mentalese. Ziff fares theories, where on the other view only two are needed. First, there is no better than Katz or Davidson in showing that we need a composi­ what we have been calling a theory of the meaning of thought, an ac­ tional theory of meaning for the system of representation that we think count of meaning for whatever system of representation one uses to with. If speakers of English think in English and we rely on that fact think with. Katz and Fodor say nothing about this theory, perhaps in communication, Ziff gives us no reason why we need a compositional because they take thoughts to be intrinsically intelligible. On the other theory of meaning for English. hand the other view takes this theory to be the most important part 5. Language used in thought. According to those who say that speak­ of the theory of meaning. Second, there is the theory that associates ers of a natural language also think in that language, there are levels sentences (under various structural descriptions) used in communica­ of meaning. A theory of the first level must account for the meaning tion with meanings they have (in isolation from setting and discourse). of an expression as a function of its role in thought. A theory of the This theory associates, e.g., sentences of English (conceived under their second level must account for the meaning of an expression used to structural descriptions) with "readings" in the system of representation communicate a thought. A theory of the second level must presup­ used in thinking. Katz and Fodor take this theory to be the central pose the first, since linguistic communication typically communicates a part of the theory of meaning, which amounts to a theory of how sen­ thought that can be expressed (roughly speaking) in the same words tences in, e.g., English are to be translated into Mentalese. The other used for communication. Such exploitation is even involved in ironic view takes this part of the theory to be trivial, on the grounds that and other nonliteral remarks, since in such cases the words of the rele­ speakers of English think in English and can use what amounts to the vant thought will be some simple function of the words in the message. homophonic scheme of translation. (I shall say more about this in a Compare this view with the one expressed in Katz and Fodor's paper, moment.) Third, there ~s the theory of the meaning of a message in a "The Structure of a Semantic Theory." 44 The theory is put forward as particular linguistic and nonlinguistic setting. Both sides argree that a level-two theory of meaning in communication. In particular it is sup- this is an important part of a theory of meaning and that it presupposes ••"On Understanding Understanding Utterances," secs. 3, 4. the other parts. "J. J. Katz and J. A. Fodor, "The Structure of a Semantic Theory," Language, How are we to decide between a view like Katz and Fodor's, which 39 ( 1963) : 170- 210; reprinted in Fodor and Katz, The Structure of Language, and in takes language to be used primarily in communication, and the alterna- Jakobovits and Miron, Readings in the Psychology of Language. Page reference is to this latter source. .. Ibid., p. 399, 288 289 Gilbert Harman LANGUAGE, THOUGHT, AND COMMUNICATION tive, which takes language to be used in thought? I tend to favor the one be able to make formal moves with that expression, at best this latter view, mainly on grounds of simplicity. But on the other side it shows merely that one understands the expression as having a particular might be argued that the former theory is needed to account for a!l grammatical form. One must also be able to give paraphrases and make the facts. Katz claims that his and Fodor's (and Postal's46 ) semantic inference that involve changes ?n nonlogical and nongrammatical vo­ theory can explain a great number of different things: cabulary. One must be able to see what sentences containing the rele­ beside requiring a semantic component ~o predict semantic ~nomaly vant expression imply, what they are equivalent with, etc. and ambiguity, we also require it to predict such other semantrc. prop­ It is here, of course, that Katz and others have imagined that appeal erties and relations as synonymy, paraphrase, antonrrny, s~manhc dis­ must be made to meaning, to entailment by virtue of meaning, to tinctness, semantic similarity, inclusion ?f senses, n~consistency, ana­ lyticity, contradiction, syntheticity, entailment, possible answer to a equivalence by virtue of meaning, etc. But that is a mistake. The rele­ question, and so forth.47 vant notions of equivalence and of implication are the ordinary ones: equivalence or implication with respect to one's background assump­ He may justly complain that we have no right to reject transformational tions, where no distinctions need be made between analytic or synthetic semantic theory unless we have some other method to explain or ex­ background assumptions. One has an understanding of an expression plain away the phenomena in question. In the remainder of this essay to the extent that one can paraphrase sentences containing it, can make I want to describe and defend one alternative method. In the process inferences involving such sentences, etc. It adds nothing to one's under­ I hope to say just what (in my opinion) is and what is not salvageable standing if one can distinguish "analytic" equivalence and implications in the views of Chomsky, Katz, and Fodor discussed above. from "synthetic" ones. In fact, most people cannot do so. Only those I shall be concerned mainly with the level-one theory of meaning, i.e., who have been "indoctrinated" can; and they are not the only ones who the theory of the meaning of language used to think in. I sh~ll say understand the language they think in.48 something about level two, which is concerned with the meamng of Here it might be objected that sentences used in thinking are often messages, only indirectly and in passing. ambiguous. How can we account for that- and for a person's ability What is it for an expression to have a meaning on level one? It is to understand or interpret a sentence one way at one time and another certainly not that the relevant person, the "thinker," can assign it one way at another time? Can we account for it without assuming that or more "readings." It is rather that he can use it in thought, i.e., that he assigns an interpretation or "reading" to the sentence in the way it has a role in his evidence-inference-action language game. He may be in which Katz suggests? Well, we can and we can't. An expression is able to use the expression in direct perceptual reports. He must cer­ ambiguous if a person can sometimes treat it as having one sort of role tainly be able to use it in (theoretical and practical) reasoning. . and at other times treat it as having a different role. Treated one way Some reasoning is relatively formal. It depends only on the logical or it admits of paraphrases it does not admit when treated the other way. grammatical form of relevant expressions and is not a function of no~­ This difference in paraphrasability represents the difference in interpreta­ logical or nongrammatical vocabulary. In order to account for a ~rson s tion; but recall that paraphrasability is relative to background knowl­ ability to reason formally, it is plausible to suppose that he thmk~ of edge and need not permit any analytic-synthetic distinction.49 the sentences he uses in thought as grammatically structured. He views To account for the difference in the ways a person can use an am­ them or conceives them as having one or another "deep structure" biguous expression we must suppose that he does not simply view it as grammatical analysis. I shall return to this point in a moment. First I a sequence of words. He views it, or hears it, as having a particular syn­ note that it is not sufficient for full understanding of an expression that tactic structure and as containing words in one or another of their pos- '"J. J. Katz and Paul M. Postal, An Integrated Theory of Linguistic Description (Cambridge, Mass.: M.l.T. Press, 1964). . ,, . .. I have discussed this issue from a slightly different point of view in "Quine on "J. J. Katz, "Recent Issues in Semantic Theory, Foundations of Language, Meaning and Existence, I," Review of Metaphysics, 21(1967-1968):124-151. 3(1967) :133. ••Ibid., pp. 150-151. 290 291 Gilbert Hannan LANCUACF:, TllOUCllT, AND COMMUNICATION sible senses. Let us consider each of these things in tum. sclected.ft0 One reason why no more is needed than a subscript is that Syntactically ambiguous sentences may be heard as having either of the relevant sorts of inference and paraphrase are those possible by vir­ two (or more) different syntactic structures. They are like the lines on tue of background information. Such background information itself paper which may be seen as a staircase viewed from the back or from must be "stored" as sentences under certain structural descriptions in­ the front. Or they are like a group of dots that may be seen as two cluding subscripted words. But a person can understand the expressions groups in one way or as two other groups in another way. Or they are he uses without having divided his background information into a part like the figure that can be seen either as a duck or as a rabbit. Thus true by "definition" (the dictionary) and another part not true by defi­ consider, "They are visiting philosophers." We may see or hear this nition (the encyclopedia). in two different ways, depending on whether we take "are visiting" We think with sentences conceived under particular structural de­ together or "visiting philosophers" together. We hear the sentence as scriptions, where we may count the subscripts that distinguish word admitting either of two groups of paraphrases, either "They are phi­ senses as part of the structural descriptions. More precisely, so-called losophers who are visiting," etc. or "They are visiting some philoso­ propositional attitudes are sentential attitudes, where the relevant sen­ phers," etc. Similarly, consider "Visiting philosophers can be unpleas­ tences are conceived under particular structural descriptions. In under­ ant." The difference in interpretation again depends on how we conceive standing what someone else says to us, we must determine the content that sentence's grammatical structure, although the difference is not of his utterance taken literally. It is sufficient that we should attend simply a matter of grouping on the surface. It is rather a matter of what to the sentence he utters conceived under the relevant structural de­ transformational theorists refer to as "deep structure." It is a matter of scription. Therefore, usually we must assign a structural description to how we conceive the source of "visiting philosophers." It may be heard his words in the sense that we must hear his words as having a particu­ as coming from "someone visits philosophers" or from "philosophers lar syntactic structure. That is not to say that we come (even uncon­ visit someone." It is true that the average person is quite ignorant of sciously) to know that it has a particular structure; and it is certainly transformational grammar. But that does not mean he fails to hear that not to say that we have knowledge of the principles that relate phonetic sentence as having one or another of the indicated structures. A person representations to structural descriptions. The situation is strictly analo­ can see lines on a page as forming one or another three dimensional gous to other cases of perceiving something as something. We can per­ structure \vithout knowing any geometry. Therefore I think that in ceive a series of lines as a particular three-dimensional structure without order to account for the way in which a person can deal with ambigu­ thinking that it has that structure, indeed without knowing anything ous sentences, we must assume that in some sense he conceives a sen­ about geometrical structure. We can certainly do so without knowing tence used in thought as having one or another syntactic structure. As rules that relate two-dimensional figures and three-dimensional struc­ I have already said above, I think that the same conclusion can be tures. reached if we attempt to account for the formal inferences a person can (To say that a speaker of a language must have unconscious knowl­ make, even if we ignore ambiguity. edge of the grammar of his language is strictly analogous to saying that Similarly, in order to account for the way in which a person deals a person who sees the world in three dimensions must have unconscious with ambiguous words, we must assume that he distinguishes a word knowledge of geometry~ To say that a speaker has internalized a gen­ used in one sense from the same word used in another sense. But we erative grammar is like saying a perceiver has internalized a geometry. can do that without assuming that he makes the distinction by assign­ Perhaps this is sometimes a harmless way of speaking. But one must be ing different readings to the word. He may mark the distinction in careful lest it illegitimately lead one to a compositional theory of mean­ sentences by a device as simple as a subscript. The inferences and para­ ing or to a rationalist theory of concept formation.) phrases that are then permissible depend in general on the subscript 00 Cf. Hannan, "What an Adequate Grammar Could Do," Foundations of Lan­ guage, 2(1966) :134-136. 292 293 Gilbert Harman LANGUAGE, TllOUCH'f, AND COMMUNICATION Sounds reach a person's ears and, after physiological processes we tically ruled out, so that they can account for a certain amount of dis­ know little about, he perceives or conceives a sentence under a particular ambiguation with their theory.52 Second, at least Katz thinks that the structural description. His understanding of that sentence is represented theory shows how certain sentences are analytic, others contradictory, not by his having assigned it a "reading" in Mentalese but rather by and stiil others synthetic, so that the theory can account for native his being able to use the sentence under that structural description in speakers' intuitive judgments about analyticity, etc.53 his thought. If we like, we may still speak of "decoding" here, in a sense But there is no such thing as semantic anomaly and no such distinc­ familiar perhaps from information theory. One decodes certain .sounds tion as the analytic-synthetic distinction. To believe otherwise is to into a sentence plus structural description. Although this point may be suffer from a lack of imagination. For example, Katz and Fodor argue partially responsible for Katz's remarks about decoding, it is not exactly as follows: what he had in mind. In the present case we have decoding of sound Now let S be the sentence The paint is silent. English speakers will into sentence under structural description. Katz speaks of decoding the at once recognize that this sentence is anomalous in some way. For phonetic representation of the sentence. He says, you will recall: example, they will distingu'ish it from such sentences as The paint is wet and The paint is yeIIow by applying to it such epithets as 'odd,' The speech sounds that stimulate these organs are then converted into 'peculiar,' and 'bizarre.' . . . Hence, another fa(:et of the semantic abil­ a neural signal from which a phonetic representation equivalent to the ity of the speaker is that of detecting semantic anomalies.54 one into which the speaker encoded his message is obtained. This repre­ sentation is decoded into a representation of the same message that the But surely, whatever anomaly there is in the phrase "silent paint" is due speaker originally chose to convey by the hearer's equivalent system of to the fact that paints do not emit noise. If some paints did and some linguistic rules.51 did not, "silent paint" would not be anomalous. And that is to say that But that is not what usually happens. One does not first perceive the there is nothing peculiarly semantic about the anomaly. Its anomaly is sounds as a sequence of words and then assign a structural description. a result of our general nonlinguistic knowledge of the world. It is a commonplace of transformation theory that one's understanding Again, Katz and Fodor argue that a theory that incorporates only the of the sentence will partially determine what words he hears. Whatever sort of considerations I have sketched above "decoding" takes place generally translates certain sounds or (perhaps) will not be able to distinguish the correct sense of seal in One of the neural signals directly into a sentence with structural description. Such oil seals in my car is leaking from such incorrect senses as 'a device bear­ a "sound-meaning correspondence" is not given simply by the rules of ing a design so made that it can impart an impression' or 'an impression grammar. made by such a device' or 'the material upon which the impression is made' or 'an ornamental or commemorative stamp' and so forth, since I conclude from the above considerations that it is possible to give all of these senses can apply to nominal occurrences of seal.55 an account of how people understand sentences that incorporates the insights gained in transformational grammar without leading one to But it requires only a little imagination to see that "seal" may have any postulate that a speaker knows that sentences have certain structural de­ of these senses, although in ordinary discourse it would be more likely 56 scriptions, that he knows the rules of grammar, or that he understands to have the sense the authors have in mind. sentences by assigning readings to them. ••Katz and Fodor, "The Structure of a Semantic Theory," pp. 408-410 . . 6. Anomaly and Synonymy. Now Katz and Fodor still have one more 03 Katz "Analyticity and Contradiction in Natural Language," in Fodor and Katz, The Str~cture of Language, pp. 530-541; The Philosophy of Language, pp. 193- card up their collective sleeve. They claim that their semantic theory 220; and "Some Remarks on Quine on Analyticity," Journal of Philosophy, 64 can account for at least two things that it would be difficult to handle ( 1967): 35-52. "'Katz and Fodor, "The Structure of a Semantic Theory," pp. 402-403. in any other way. First they think their theory can show how certain '"'Ibid., p. 409 . interpretations of a sentence that are grammatically possible are seman- .. For further discussion of the question whether it is possible to distinguish semantic anomaly or disambiguation from anomaly or disambiguation due to our 01 Katz, The Philosophy of Language, p. 104; my emphasis. generally accepted beliefs, see Dwight Bolinger, "The Atomization of Meaning," 294 295 Gilbert Harman l.ANCUACF:, TUOUCHT, AND COMMUNICATION McCawley has argued a similar point with respect to purported syn­ Second, the intuitive distinction Katz and others make between ana­ tactic anomalies of a certain type: lytic and synthetic truths is easily explained away without appeal to Moreover, it appears incorrect to regard many so-called "selectional vio­ transformational semantic theory: people who have such intuitions suf­ lations" as not corresponding to possible messages, since many of them fer fro~ a lack of imagination. The intuitions come from their inability can turn up in reports of dreams: to imagine that certain sentences are false. But as many philosophers l ( 2) dreamed that my toothbrush was pregnant. have pointed out, after a little practice such things can be imagined.59 ( 3) I dreamed that I poured my mother into an inkwell. ( 4) I dreamed that I was a proton and fell in love with a shapely 7. Final remarks. Consider the two views. On the first view, a person green-and-orange-striped electron. who speaks a natural language can think in that language and does not or in reports of the beliefs of other persons: need to translate sentences conceived into some other system of repre­ sentation, Mentalese. On this view, Mentalese incorporates one's natu­ ( 5) John th~nks that el.ectrons are green with orange stripes. (6) John thmks that his toothbrush is trying to kill him. ral language. On the second view, a person cannot think in language (7) John thi~ks that ideas are physical objects and are green with and must translate sentences of his language into Mentalese. The in­ orange stnpes. corporation view has the following advantages over the translation view: or in the speech of psychotics. While one might suggest that a paranoid (I) The incorporation view provides a natural explanation of the way who says things like in which learning one's first language makes possible thoughts and other (8) My toothbrush ij; alive and is trying to kill me. propositional attitudes one would not otherwise have. The translation has differe~t selectional restrictions from a normal person, it is pointless view must invoke some special principle to account for this. (2) Simi­ to do so, smce the difference in "selectional restriction" will correspond larly, the incorporation view suggests an explanation of the way in ex~ctly to a difference in beliefs as to one's relationship with inanimate which unconscious thinking makes use of verbal punning, as revealed in ob1ects; a per~on ~~o utters sentences such as (8) should be referred psychoanalytic studies of dreaming, slips of the tongue, and free asso­ to a psych1atnc clime, not to a remedial English course.57 ciation.60 It is not clear how the translation view could explain such I agree and think that supports the view that there is no real distinction things. (3) The translation view is unnecessarily complicated because between semantic anomaly and anomaly due to "extralinguistic factors." it cannot explain anything that is not explicable on the incorporation Similar points apply to Katz's claims about analyticity, etc. First, the view. ( 4) The translation view suggests things that are false about fact that people have "intuitions" about analyticity shows at best that anomaly, synonymy, etc. there is a distinction between "seems analytic to certain people" and Such considerations seem to be the relevant ones to use in deciding "seems synthetic to them." It does not show that there are sentences which of the two views is more correct. For example, it is difficult to that are really analytic as opposed to others that are synthetic. The fact see how neurophysiological evidence could support one of the views that people once had "intuitions" that some women were witches and against the other (except by pointing to further relatively behavioral others not, certainly fails to show that there were women that really phenomena to be explained). For, as descriptions of the mechanism of were witches as opposed to others that were not.58 the brain, the two views must be taken to be descriptions of a fairly Language, 41(1965):555-573, reprinted in Jakobovits and Miron, Readings in the '"'See e.g., Hilary Putnam, "It Ain't Necessarily So," Journal of Philosophy, Psychology of Language, pp. 432-448. 117 59(1962) :660; J. M. E. Moravcsic, "The Analytic and the Nonempirical," Journal James D. McCawley, "Where. Do .Noun ~hrases Come From?" in R. A. Jacobs of Philosophy, 62(1965) :421-423; Harman, "Quine on Meaning and Existence, I." and P. S. Rosenbaum, eds., Readmgs m English Transformational Grammar (Wal­ Quine speculates on the mechanism of analyticity "intuitions" in Word and Ob­ tham, Mass.: Ginn, 1970), p. 168. ject, pp. 56-57, 66-67 . .. Cf. N. L. Wilson, "Linguistic Butter and Philosophical Parsnips" p. 65· Har- 00 I owe this point to Lucy Harman. man, "Quine on Meaning and Existence, I," pp. 137-138. ' ' 297 296 Gilbert Harman - - ---NOAM CHOMSKY----- abstract sort. And it is not easy to see how any neurological mechanism that might account for the relevant behavior and that could be inter­ preted as an instantiation of one of the descriptions could not also be interpreted as an instantiation of the other. Therefore, I see no reason not to accept the incorporation view.s1 Knowledge of Language

01 """ I discuss these issues further in Thought (Princeton, N.J.: Princeton University Press, 1973).

There are a number of different questions that I would like to touch upon in these lectures, questions that arise at various levels of generality and tha.t grow out of different, though not totally unrelated concerns. I want to present a certain framework within which, I believe, the study of language can be undertaken in a very fruitful way - not the only framework, to be sure, and one better suited to certain problems than to other; equally legitimate ones. Within this framework I would like to discuss some technical questions that are at or near the borders of research. At this level of discussion, I will be presenting some material that is internal to the theory of transformational generative grammar. But I would also like to suggest that this rather technical material is potentially of quite general interest, that by studying it we can hope to learn some important things about the nature of human intelligence and the products of human intelligence, and the specific mechanisms that enable us to acquire knowledge from experience, specific mecha­ nisms that, futhermore, provide a certain structure and organization for, and no doubt certain limits and constraints on, human knowledge and systems of belief. I think that the work I will describe at least hints at a concept of man that is rather different, in interesting respects, from others that have been implicit in much modern thinking about these matters, and would like to elaborate on this question as well. I have in mind, then, a large enterprise, of which only a small part can be carried out with a satisfactory degree of clarity and precision. Still, I think it is useful to consider this small part against the background of what might ultimately be achieved. The study of language, as I will be considering it here, can be regard­ ed as a part of human psychology. It forms a part of the general study

AUTHOR'S NOTE : This essay is the first of six John Locke Lectures delivered at Oxford University in May-June 1969. Parts appeared in the London Times Literary Supplement, May 15, 1969. The other lectures have not yet been published. 298 299